on the subject of

racial bias in generative AI

Creating a self-portrait is a personal and, by definition, introspective process. During my initial back and fourths with MidJourney, I was excited to see how the AI would interpret my features and create a portrait that was uniquely me. During this exercise, I came across some disheartening realizations.

“professional” = white

As you may know, MidJourney is able to take an image along with a set of chosen keyword prompts and spit out a newly generated image based on that input. I noticed that when using my headshot (pictured) along with any keyword prompts that included the word “professional”, the output image from MidJourney would always look like a white dude (also pictured). I had to either leave out the word professional, and sometimes even “headshot”, and specifically include “black guy” as a keyword to make sure the output looked something like me.

This experience opened my eyes to the broader issue of racial bias in AI, and the need for more inclusive and diverse data sets to ensure that AI accurately represents all individuals. Just add this to the long list of random racial BS we deal with on a daily basis that often goes unnoticed by those less melanated individuals. Thankfully, it seems that these types of dataset issues and embedded racial biases are actively being worked on - I am hopeful that a more equitable future lies ahead in the AI / technology space.